FaaS – Hyped up new acronym or a way to level the simulation playing field? The opportunities and challenges of FaaS (Function as a Service) for accelerating simulation and analysis.
DescriptionSaaS (Software as a Service) and HPC (High Performance Computing) are now common terms in the design community. However, an emerging concept is that of FaaS (Function as a Service). FaaS holds promise for drastically improving circuit, electromagnetic, and signal integrity simulation performance while enabling rapid analysis and design exploration. Yet adoption of FaaS in semiconductor, printed circuit and RF design roles has been slow, and many organizations have questions and concerns about its feasibility, security, or are simply unaware of its existence.
The design community traditionally invests in large server infrastructure when there's a need to accelerate circuit simulation, SI, and electromagnetic modeling. In addition to the heavy cost of hardware and power consumption, configuration management and maintenance costs can often outweigh the benefits of data security and optimization. Moreover, companies that deploy HPC internally become committed to running on an "upgrade-or-die” hamster wheel.
FaaS is an attractive alternative because it allows vendors to execute small pieces of code on the network edge, making the codebase more scalable without having to spend resources on maintaining the back-end. With FaaS, provisioning resources takes a few seconds rather than minutes or hours and is not limited to a particular language or resource. FaaS is fast becoming a method of choice for accelerating compute intensive applications that retain a desktop environment such as EDA tools. This approach, known as "microservices,” is appealing because it also means vendors can easily make incremental improvements and new features to back-end code without costly desktop deployments.
With traditional SaaS HPC, when a function has not been called for a period of time, the provider shuts it down to save energy and avoid over-provisioning. The next time a user runs an application that calls that function, the provider spins it up fresh and starts hosting that function again, which takes time. This startup latency, known as a "cold start,” is one of several bottlenecks to adoption.
In contrast, FaaS worker node functions spin up at the edge in a very short amount of time. As more such drawbacks are addressed and the popularity of edge computing grows, we can expect FaaS adoption to grow in the EDA and Test/Measurement landscape. But will it?
FaaS decouples the user from back-end computation in a way that offers unprecedented scalability, but how is it different from the current generation of EDA HPC cloud acceleration options? And how can it be leveraged to save time and cost during development? Why would one choose FaaS over owned internal HPC resources?
This panel will bring together experts from semiconductor design, EDA software, and electronic product integration to discuss the opportunities and the drawbacks – perceived and real – of using FaaS for acceleration of design and simulation workflows. The diverse group on the panel will ensure a broad set of interests and viewpoints are represented.
DAC Pavilion Panel
TimeMonday, July 10th2:00pm - 2:45pm PDT
LocationDAC Pavilion, Level 2 Exhibit Hall